Moschella et al. (2022)

Demonstrated zero-shot model stitching across text models trained on different modalities

Neighborhood — ranked by edge-count

paper

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Merullo et al. (2022)concept0.777
Found a single linear projection sufficient to stitch vision model to LLM for VQA and captioning
Zimmermann et al. (2021)concept0.747
Showed contrastive learning inverts the data generating process; supports claim that contrastive learners recover statistics of underlying world
Sorscher et al. (2022)concept0.737
Showed word embeddings of visual concept names can be isometrically mapped to image embeddings, and developed framework for efficient concept extraction
Dravid et al. (2023)concept0.720
Found Rosetta Neurons — neurons activated by same patterns across diverse vision models
Abdou et al. (2021)concept0.711
Discovered that color distances in language representations mirror human perceptual distances; reproduced and extended in this paper
Abrini et al. 2025concept0.706
Recent ToM in AI work cited as emphasizing inference of others' beliefs, contrasted with this paper's self-focus
Baker et al. 2017concept0.699
Reference for ToM modeled through partially observable inference of others' beliefs
Navratilova et al. 2012concept0.691
Cited to support the relief-seeking interpretation of chronic agent food-discovery behavior